January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

COMPLIANT MOTION CONSTRAINTS

CHRISTOPH M. HOFFMANN AND WEIQIANG YANG Department of Computer Science Purdue University West Lafayette, IN 47907, USA

We examine a dual-quaternion formulation for expressing the relative rigid body motion between two objects when incidence constraints are to be observed. The incidences are between points, lines and planes, of the two parts. Both parametric and implicit representations are investigated. Several examples illustrate the techniques. Keywords: Rigid body motion, compliant motion, quaternion, dual quaternion, relative motion, incidence constraint, virtual reality, kinematics, geometric constraint.

1. Introduction Geometric constraints are used in two dierent contexts. In one application area we dene a set of geometric primitives and constraints upon them, and then are asked to nd an arrangement of the primitives such that the constraints are satised. Let us call this the construction problem. The construction problem arises for example when dening CAD models for discrete manufacturing. In a second application area we are given a set of (usually composite) geometric objects as well as constraints upon their spatial relationship, with the objective of constraining the relative motion of the objects with respect to each other. Let us call this the compliance problem. The compliance problem arises in assembly modeling, kinematic simulation of machinery, and in virtual reality, to name a few uses. In this paper we consider the compliance problem in 3-space and investigate basic techniques to solve it. There is a wealth of prior work,
Work

supported in part by NSF Grants EIA 02-16131, DMS/CCR 01-38098, CCR 99-02025, and by ARO contract 39136-MA. 1

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

and we give a few example references below. The area of kinematics to a large extent considers and solves compliance problems, for instance when investigating linkages, or, more generally, when designing and analyzing machinery [6,7]. Other relevant research is done in robotics [1], and in some areas of geometric constructions [5]. To some extent, the compliance problem overlaps with the construction problem, as seen in [4], where a system of equations is attacked by considering the residual compliant motion of geometric primitives when restricting to a subset of the given constraints. Much of the research into compliance is dominated by seeking elegant mathematical formalisms that would simplify expressing and analyzing compliant motion. In addition to ad-hoc techniques that are highly successful in special cases such as four-bar linkages, three main formalisms have emerged: (4 4) transforms, screws, and dual quaternions. The three formalisms oer a general description of rigid body motion in 3-space. Note that screws are essentially dual quaternions, but the reduced coordinate set may introduce ambiguities in some cases. For this reason we do not consider them further. 2. Tools and Notation In this section, we review some basics and notations on quaternions, their relations to rotations, and dual quaternions. 2.1. Quaternions The eld of quaternions has elements of the form a = a0 + a1 i + a2 j + a3 k where the coecients ar are real numbers and the units i, j, and k obey the equations: i2 = j 2 = k 2 = 1, ij = ji = k, jk = kj = i, ki = ik = j

Complex numbers are quaternions with a2 = a3 = 0. The length of a quaternion a, is dened as a = a2 + a2 + a2 + a2 . Quaternions are 0 1 2 3 due to Maxwell. The conjugate of the quaternion a = a0 +a1 i+a2 j+a3 k is the quaternion a = a0 a1 i a2 j a3 k. The norm of a is the square of the length of a and is equal to the quaternion product aa. We dene the inner product (a b) of two quaternions a and b as (a b) = a0 b0 + a1 b1 + a2 b2 + a3 b3 . Note that the norm of a is (a a) = aa.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

A quaternion with a0 = 0 is called a vector. More generally, we can call a0 the real part of the quaternion a and call a1 i + a2 j + a3 k the vector part. We denote quaternions with lower-case bold letters. Vectors in R3 are denoted by bold lower-case letters with an arrow; for example, p. The arrow is omitted when it is clear from the context that we speak of a vector. The vector part of a quaternion is denoted in the same way. Thus, if a = a0 +a1 i+a2 j +a3 k, then a = a1 i+a2 j +a3 k. We note that a = a0 a. We will use the inner product, denoted by and the cross product of vectors, denoted by , to express quaternion operations more succinctly. For example, the product of two quaternions a = a0 + a and b = b0 + b is the quaternion ab = a0 b0 (a b) + a0 b + b0 a + a b 2.2. Rotations With Cartesian point coordinates in 3-space, a rotation in 3-space about the origin can be represented by the orthogonal matrix r11 r12 r13 R = r21 r22 r23 r31 r32 r33 where RRT = I and det(R) = 1. It is well-known that unit-length quaternions can represent rotations about the origin. Wittenburg [10] gives the following conversion formulae. For any unit-length quaternion a, the entries of the rotation matrix are r11 = 2(a2 + a2 ) 1, 1 0 r21 = 2(a1 a2 a0 a3 ), r12 = 2(a1 a2 + a0 a3 ), r13 = 2(a1 a3 a0 a2 ), r22 = 2(a2 + a2 ) 1, 2 0 r13 = 2(a2 a3 + a0 a1 ), r33 = 2(a2 + a2 ) 1, 0 3

r31 = 2(a1 a3 + a0 a2 ), r32 = 2(a2 a3 a0 a1 ),

and for any rotation matrix with entries rpq the quaternion coecients are: a2 = (r11 + r22 + r33 + 1)/4, 0 a2 = r11 /2 u, 1 a2 = r22 /2 u, 2 a2 = r33 /2 u, 3 where u = (r11 + r22 + r33 1)/4. Other, equivalent conversion formulae are given in, for example, [8].

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

There is a well-known geometric interpretation of the quaternion representation of such rotations. Let v = (v1 , v2 , v3 ) be the unit-length direction vector of the axis of rotation, and let 2 be the angle of rotation. With c = cos() and s = sin(), the rotation is represented by the quaternion c + sv1 i + sv2 j + sv3 k. 2.3. Dual Numbers and Dual Quaternions A dual number is dened as A = a + b , where a and b are from a eld and 2 = 0. Dual numbers form a Cliord algebra. If A = a + b is a dual number, A = a b is its conjugate. A dual quaternion is dened as A = a + b , where a and b are quaternions. Equivalently, a dual quaternion is a quaternion whose components are dual numbers (with real coecients). Dual quaternions can represent points, lines and planes in 3-space, as well as general rigid body motions, as will be discussed in the next section. As in [2], we dene three dierent conjugations of a dual quaternion, according to whether the quaternion components are conjugated, the dual numbers are conjugated, or both. Let A = a + b , where a and b are the quaternions. We dene A=a+b , A =ab , A = ab .

3. Representations We adopt the algebraic schema of [2] to represent points, lines and planes in 3-space, as well as rigid-body transformations on them. 3.1. Points, Lines, and Planes In Cartesian coordinates, points are specied by their position vector (p1 , p2 , p3 ) which we represent by the dual quaternion P=1+p A plane has the equation ap1 + bp2 + cp3 + d = 0, where we require that a2 + b2 + c2 = 1. Such a plane is represented by the dual quaternion E=n+d . The rst quaternion is the plane normal vector n = ai + bj + ck, and the second quaternion, which is real, is the constant of the implicit plane equation.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

Using Plcker coordinates, lines in 3-space can be represented by two u 3-vectors t = (t1 , t2 , t3 ) and m = (m1 , m2 , m3 ), where t is the line direction vector, normed to unit length, and m is the moment vector p t of some point p on the line. Clearly, the inner product of the moment vector and the direction vector is zero; that is, (m t) = 0. Identifying, as before, the vector (a, b, c) with the quaternion ai + bj + ck, we represent the line (t, m) as the dual quaternion L=t+m . The rst quaternion is the unit-length direction vector, the second quaternion is the moment vector. For lines through the origin m = 0. 3.2. Rigid Body Motion The unit quaternion q was noted to represent a rotation about the origin. The dual quaternion Q = q with the zero quaternion as the coordinate is chosen to represent the same rotation. Furthermore, we represent a translation by the vector (2s1 , 2s2 , 2s3 ) by the dual quaternion S = 1 + s , where s = s1 i + s2 j + s3 k. A rigid body motion in 3-space can therefore be represented by a dual quaternion T = SQ that is the product of the rotation quaternion Q and the translation quaternion S, imitating the action of 4by-4 transforms. The representation of rigid motions by dual quaternions is due to Study [9]. Screw Motion Chasles theorem [3] states that every rigid motion is equivalent to a screw motion. Here, the screw with axis (t, m), angle of rotation 2, and a displacement 2d is represented as the dual quaternion Mscrew = cos() + sin() t + (d sin() + sin() m + d cos() t) (3.1)

Note that for = 0, the motion M simplies to 1 + d t , a translation by 2d t, and for d = 0 and m = 0 it simplies to cos() + sin() t, a rotation about the origin. Other Motion Representation The general rigid body motion can be expressed as M=q+u , We prove that this is true. q = 1, (q u) = 0. (3.2)

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

The conditions of Equation (3.2) are clearly satised by the screw motions of Equation 3.1. Thus, all rigid motions can be represented. Conversely, assume that the above conditions are satised by the dual quaternion M = q + u . Writing q = q0 + q1 i + q2 j + q3 k, we may dene , t1 , t2 , t3 by setting q0 = cos(), and qr = sin()tr , r = 1, 2, 3. Let t = t1 i + t2 j + t3 k. Since q = 1, we have t = 1 in general. If |q0 | = 1, then q1 = q2 = q3 = 0 and u0 = 0. In that case M is a translation. If u0 = 0, the vector u must be perpendicular to the vector q. We may assume |q0 | = 1 and dene the vector m = u/ sin(). We now see that M represents a pure rotation about an axis with direction t and moment m. Otherwise, with |q0 | = 1 and u0 = 0, we have sin() = 0, and we can dene the nonzero quantity d from u0 = d sin(). Dene m0 = 0 and mr sin() = ur dtr cos(), r = 1, 2, 3. Then 0 = q0 u0 + q1 u1 + q2 u2 + q3 u3 = d sin() cos() + = d sin() cos() + = sin2 ()(t m) Therefore the vector m = m1 i + m2 j + m3 k is perpendicular to the vector t, which means that M is a screw motion with axis (t, m), rotation angle 2, and displacement 2d. 3.3. Motion of Points, Lines and Planes Let P be a dual quaternion representing a point or a plane. Then the dual quaternion P that represents the result of a rigid body motion M, applied to the point or plane represented by P, is calculated as P = MPM . (3.3) Similarly, the line represented by the dual quaternion L is transformed into the line represented by L , where L = MLM (3.4) An algebraic computation veries this denition; see also [2]. Summarizing, dual quaternions allow us to represent points, lines and planes in 3-space uniformly, and express rigid body transformations of them uniformly as well.
3 r=1 sin()tr (mr sin() + dtr cos()) 3 2 2 r=1 (d sin() cos()tr + tr mr sin ())

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

4. Constrained Motion We investigate what relative motion is possible when requiring a single incidence constraint of a point, line or plane on another point, line or plane. First, we formulate incidence conditions in terms of dual quaternions. Then we investigate relative motion assuming that the incidences are currently satised. 4.1. Incidence Six elementary incidence conditions arise when requiring points, lines and planes to be incident to each other. Among features of equal type, incidence is trivial, as it requires equal coordinates. Note, however, that for planes and lines incidence with opposite orientation should be accounted for. Let P be a point, E a plane, and L a line. We require that the plane normal and the line direction vectors have unit length. The following are the incidence conditions between features of dierent type. EP + PE = 0 LP PL = 0 LP + PL = 0 See [2] for a proof. 4.2. Parametric Relative Motion It is not dicult to express parametrically the relative motion that obeys a single elementary incidence constraint. In particular, if the elements are of the same type, we are asking for motion expressions that leave a point, a line or a plane invariant. However, a parametric representation in the presence of multiple incidence constraints between dierent features of two rigid bodies is not so easy. We will show that it can be done based on the parametric representation, in a number of cases. A related problem commonly investigated in robotics is to synthesize the motion of a kinematic chain, such as an articulated robotic arm. Such work typically assumes xed common lower-pair connections between the links such as a revolute or a prismatic joint. If we express the relative motion of a single incidence constraint parametrically, then we can combine the equations into a single system and obtain a combined parameterization using elimination computations. It is advantageous to keep the equation system as simple as possible, and this point on plane point on line line on plane

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

would argue for performing every algebraic simplication possible as a preprocessing step before undertaking the actual evaluation. We begin with expressing the elementary constraints. As before, we denote points, lines and planes with dual quaternions P, L and E, respectively. 4.2.1. Incidences of Equal Type These are incidences of point on point, line on line, and plane on plane. To express the relative motions, we ask which rigid body transformations x a point, a line or a plane. For the point represented by the dual quaternion P = 1 + p we obtain MP = q + p q, q = 1. (4.1) Note that we require that q has length 1, and that p is a vector quaternion, that is, p0 = 0. We can derive MP by conjugating a general rotation about the origin by the translation of the xed point to the origin. Let T be the translation from the point P to the origin, represented as a dual quaternion, and let its inverse be T . Then MP = T QT. The representation has four parameters which reduce to three independent ones because of the unit-length requirement on q. A dierent parameterization derivation is possible by considering a screw motion that has a zero displacement along the axis. With t an arbitrary unit length vector, we then obtain the equivalent form MP = cos() + t sin() + m sin(), t = 1, m = p t. Again, there are 4 parameters reducing to three independent ones because of the unit-length requirement. The resulting parameterization is identical to (4.1). Next, we consider the motion that leaves the plane E = n+d invariant. Here, n is the unit-length normal vector of the plane. The motion that leaves the plane invariant can be considered as a rotation about an axis through the origin in the direction n plus a translation by a vector t in the plane, which therefore satises (t n) = 0. We obtain ME = cos() + sin()n + (cos()t + sin()(t n)), (t n) = 0. (4.2)

The four parameters reduce to three independent ones by the (linear) equation (t n) = 0.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

Finally, the motion that leaves the line L with direction t and moment m invariant is given by ML = cos() + sin()t + (d sin() + sin()m + d cos()t), (t m) = 0. (4.3)

It represents the screw with axis L, displacement 2d, and angle of rotation 2. The two parameters d and are independent. Thus, this relative motion parameterization is irredundant. 4.2.2. Incidences of Dierent Types Unequal type incidence constraints may be obtained by combining motions that include xing one of the features, to account for symmetries, followed by displacing it within the geometry of the other feature. The relative motion subject to requiring that the point P stay in the plane E can be obtained by composing the relative motion that xes P with a subsequent translation in the plane. With 2s the vector of translation, we obtain MPE = TMP = q + ((s q) + q0 s + p q + s q) = q + (sq + p q) , q = 1, (s n) = 0. (4.4)

The condition that the translation vector be perpendicular to the plane normal implies two independent parameters in the choice of s, bringing the total degrees of freedom of the motion to 5. Applying the same procedure we obtain the following for keeping the point P on the line L. The translation must be along the line direction, thus we obtain MPL = TMP = q + (tq + t q) . (4.5)

Finally, consider keeping a line L = t+m incident to a plane E = n+d . Geometrically, the motion can be considered a screw motion with axis L followed by a translation of the line in the plane which can be restricted to a displacement perpendicular to the line. Since t is perpendicular to the plane normal, the subsequent translation is in the direction t n.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

10

We obtain the following representation: MLE = cos() + sin()t + (d1 sin() + sin()m)+ (d1 cos()t + d2 cos()u + d2 sin()n), u = t n, t = 1, (m t) = 0. (4.6)

Here 2 is the angle of rotation, 2d1 the displacement in the t direction, and d2 the displacement in the perpendicular u = t n direction. 5. Combining Constraints 5.1. Parametric Approach Consider now moving a part A relative to another part B where there are multiple incidence constraints between features of the two parts. The parametric representations of the relative motion can be used when combining several incidence constraints as follows. Let F1 , ..., Fr be the parametric forms of the residual motion taken separately for each incidence constraints. By equating the rigid body motions of the Fi , we obtain a system E1 , ..., Es of implicit equations in the parameters. We solve this system for a set of independent parameters. This is an elimination computation and therefore potentially expensive. Then we can evaluate relative motion by evaluating the dependent parameters as necessary and substituting into F1 , thus obtaining an admissible relative motion. Example 5.1. Consider a xed part A with two plane features, E1 = j and E2 = i, namely, the planes x = 0 and y = 0. On a moving part B we x the points P1 = 1 + i and P2 = 1 + j , that is, the points (1, 0, 0) and (0, 1, 0), respectively. Evidently P1 is on E1 and P2 is on E2 . We use for the translation in E1 , the vector s = (s, 0, t) and for the translation in E2 , the vector s = (0, s , t ). Then the parametric forms for the relative motion, considering the incidence constraints separately, are, in detail, F1 : M1 = q + F2 : M2 = q + sq1 tq3 + (sq0 tq2 )i s q2 t q3 + (s q3 t q2 + q3 )i

+(tq1 sq3 q3 )j + (sq2 + tq0 + q2 )k , +(s q0 + t q1 )j + (t q0 s q1 q1 )k . We equate the parameters q and q , and determine the relationships between the other parameters by equating the components of the quaternion.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

11

Accounting for q = 1, we obtain

2 2 s = q0 q3 q1 q2 q2 q3 , 2 2 s = q0 q3 q1 q1 q2 q3 ,

t = t + q0 q1 + q0 q2 q1 q3 + q2 q3 . Thus, we have four independent parameters. Three of them specify q and this determines s and s as well. The fourth parameter is t which, in conjunction with q, determines t .

Figure 1.

Conguration of Example 2; drawing reproduced from [6]

Example 5.2. Consider the joint constructed by tting a tripod of balls into three slots whose center planes intersect in a common line. See Figure 1 above. Here we have three point/plane incidence constraints. The features of the xed part are the three planes E1 = j,

3 1 i j, E2 = 2 2 3 1 E3 = i j. 2 2

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

12

The respective parametric tangential motions are s1 = (s, 0, t), s2 = (s , 3s , t ), s3 = (s , 3s , t ). The features of the moving part are P1 = 1 + i, 1 3 P2 = 1 + ( i + j), 2 2 1 3 P3 = 1 + ( i j). 2 2

The three transformations obtained are F1 : M1 = q + (sq1 tq3 + (sq0 tq2 )i + (tq1 sq3 q3 )j + (sq2 + tq0 + q2 )k), F2 : M2 = q + (s (q1 3q2 ) t q3 + + (s (q0 + 3q3 ) t q2 + 3q3 /2)i + (s ( 3q0 + q3 /2) + t q1 + q3 /2)j + (s ( 3q1 q2 ) + t q0 q2 /2 3q1 /2)k), F3 : M3 = q + (s (q1 + 3q2 ) t q3 + + (s (q0 3q3 ) t q2 3q3 /2)i + (s ( 3q0 + q3 /2) + t q1 + q3 /2)j + (s ( 3q1 q2 ) + t q0 + 3q1 /2 q2 /2)k). We equate the coordinates of the three dual quaternion expressions. Note that this results in a linear system of equations in the parameters s, s , s , t, t , t . The equations yield the following
2 2 2 2s = q1 3q2 2q3 , 2 2 2s = q1 3q1 q2 q3 , 2 2 2s = q1 + 3q1 q2 q3 , 2(t t ) = 3q0 q2 + 3q1 q3 3(q0 q1 + q2 q3 ), 2(t t ) = 3q0 q2 + 3q1 q3 + 3(q0 q1 + q2 q3 ),

q0 q3 = 0.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

13

It follows from these six equations that the relative motion has three degrees of freedom.

5.2. Implicit Approach In the implicit approach we express the relative motion as relations on the parameters of the general rigid-body motion. This will allow us to conjoin the parameter relations without having to resort to algebraic elimination.

Implicit Incidence Constraints, Equal Types We translate the parametric formulations of the incidence constraints into implicit form. We will work with the generic rigid body transformation expression M of Equation (3.2). We derived the parametric form of Equation (4.1) for keeping a point p invariant. Accordingly, the implicit conditions on M are Mp = q + u , p q = u, q = 1. (5.3)

These conditions imply in particular that u0 = 0. When p = 0 the point is at the origin and the condition on u simplies to u0 = u1 = u2 = u3 = 0. To x the plane E, we derived the parametric form of Equation (4.2). It implies the condition u0 = 0. The direction of the rotation axis implied 2 by q has to be normal to the plane, hence we require (n q) = 1 q0 . Since both t and t n in Equation (4.2) are perpendicular to the plane normal, we obtain the following conditions: Me = q + u , u0 = 0, (n q) =
2 1 q0 ,

(n u) = 0.

(5.4)

The second condition degenerates when the motion is a pure translation since, in that case, the right-hand side vanishes. However, in that case the condition q = 1 forces q1 = q2 = q3 = 0, so a pure translation within the plane is implied by the formulation. Now consider a line L = t + m with direction t and moment vector m. The line is invariant under M = q + u if the transformed line L = MLM has the same tangent and moment vectors. This implies the following relations, in which m and t are known quaternions: Ml = q + u , t = qtq, m = utq + qmq + qtu.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

14

Implicit Incidence Constraints, Dierent Types Consider now keeping a point P on a plane E, for which we derived the parametric form of Equation (4.4). We derive the implicit condition on M by requiring that the transformed point P is again in the plane E. Let P = 1 + p. We obtain P = (q + u)(1 + p)(q u) = 1 + (uq qu + qpq) = 1 + (2(u q) 2u0 q + 2q0 u + qpq) = 1+ p. An algebraic computation veries that the real component of p is zero, that is, p is the position vector of the transformed point. Assuming the original point is in the plane with unit normal vector n, we obtain the condition (n p) = (n p ), or equivalently: (n p) = 2(n (u q) 2u0 (n q) + 2q0 (n u) + (n (qpq)) (5.5)

Example 5.6. Consider the plane E1 = j and the point P1 = 1 + i in the plane. Any motion M that keeps this point in the plane must satisfy according to Equation (5.5) q0 q3 + q1 q2 u0 q2 + u1 q3 + u2 q0 u3 q1 = 0 q =1 (q u) = 0 Example 5.7. We consider the points and planes of Example 1. The conditions from the point incidences are then q0 q3 + q1 q2 u0 q2 + u1 q3 + u2 q0 u3 q1 = 0 q0 q3 + q1 q2 u0 q1 + u1 q0 u2 q3 + u3 q2 = 0 q =1 (q u) = 0 Note that a translation in the z-direction, 1+ dk, satises these conditions.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

15

Example 5.8. We consider the planes and points of Example 2. The incidence conditions, after some simplication, dene the equations q =1 (q u) = 0 2q0 q3 q1 q2 = 0 q0 q3 + q1 q2 + u2 q0 u0 q2 + u1 q3 u3 q1 = 0 2 2 2 3(u0 q1 u1 q0 + u2 q3 u3 q2 ) + 3(q1 q2 ) + 2(q0 q3 + q1 q2 ) = 0 6. Discussion The uniformity of the representation and the algebraic nature of the representation are the main attractions when using dual quaternions. Points, lines and planes are simple to represent as dual quaternions, and so are rigid-body motions. Moreover, as we have seen, there is considerable geometric intuition in this representation schema. Another advantage of dual quaternions, from a computational perspective, is that they describe a general rigid-body motion with only eight parameters, whereas a 4 4 matrix representation would require twelve. Thus, the system of equations describing a particular contact conguration is smaller. A screw representation would lower this to six parameters, but the resulting equations may fail in particular instances and do not dier, in essence, from the dual quaternion representation. There are some drawbacks to using dual quaternions as well. In the implicit form of the constraint encoding, for instance, the conditions can become fairly complex. An example is the implicit representation of motions that keep a line invariant. Here, the parametric form does better. Moreover, the implicit form we derived has some redundancies. Consider again all conditions on Ml : t = qtq m = utq + qmq + qtu q =1 (q u) = 0 (6.1) (6.2) (6.3) (6.4)

Conditions (6.1) and (6.2) each yields three scalar equations, giving eight equations in eight variables total. Therefore, there must be two redundant equations. With a pure translation (|q0 | = 1), Condition (6.1) is trivial. With a pure rotation (u0 = 0 and u = sin()m), on the other hand, Con-

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

16

dition (6.1) is not trivial. Thus, redundancy depends on the parameter values. Parametric motion representations often lead to motion descriptions in which the system of parameter relations is linear, a computational plus. However, nonlinear relations may ensue, for example on the q coordinates of the transformation. Here, symbolic algebraic computations may require reasoning that is not entirely automated in, for instance, Maple. Another drawback is that the representation of the relative motion may not include certain special cases. For instance, given a plane E, we may choose a line (t, m) that lies in the plane and use it as the axis for a rigid-body motion that has a rotation angle of 180 . Those motions also preserve the plane, albeit with a reversal of the plane orientation. Thus, the geometry and the algebra diverge in this case. In contrast to the parametric expression of relative motion, the implicit formulation does not require intermediate positions to satisfy the constraints. For example, the point P is required to be on the plane E only at the start and at the end of the motion. As the motion progresses, it may very well leave the plane E at the other times. This is true in particular of the special case of a rotation by 180 about an axis in the plane E. From a computational perspective, dual quaternions do not reduce the number of arithmetic operations that must be done to compute the image of a feature under a given transformation. Using a 44 matrix representation, transforming a point requires 12 multiplications and 9 additions. The dual quaternion representation, on the other hand, requires more. References
1. S. Ahlers and J. M. McCarthy. The Cliord algebra of double quaternions and the optimization of TS robot design. In E. Bayro and G. Sobczyk, editors, Applications of Cliord Algebras in Computer Science and Engineering. Birkhuser Verlag, 2000. a 2. W. Blaschke. Kinematik und Quaternionen. VEB Verlag der Wissenschaften, Berlin, Germany, 1960. 3. M. Chasles. Note sur les proprits gnrales du syst`me de deux corps... ee e e e Bull. des Sciences Mathmatiques, Astronomiques, Physiques et Chimiques, e 14:321326, 1830. 4. X.-S. Gao, C. Homann, and W. Yang. Solving spatial basic geometric constraint congurations with locus intersection. In Proc. 7th ACM Symp. Solid Modeling and Applic., pages 95104. ACM Press, 2002. 5. X. S. Gao, C. C. Zhu, and Y. Huang. Building dynamic mathematical models with Geometry Expert. In W. C. Yang, editor, 3rd Asian Techn. Conf. in Mathematics, pages 216224, New York, 1998. Springer.

January 5, 2003

17:33

WSPC/Trim Size: 9in x 6in for Proceedings

ch4edita

17

6. J. Phillips. Freedom in Machinery, Vol I: Introducing Screw Theory. Cambridge University Press, Cambridge, UK, 1984. 7. J. Phillips. Freedom in Machinery, Vol II: Screw Theory Exemplied. Cambridge University Press, Cambridge, UK, 1990. 8. H. Pottmann and J. Wallner. Computational Line Geometry. Springer, Heidelberg, Germany, 2001. 9. E. Study. Die Geometrie der Dynamen. Jahresbericht der Deutschen Mathematiker-Vereinigung, 8:204216, 1899. 10. J. Wittenburg. Dynamics of Systems of Rigid Bodies. B. G. Teubner, Stuttgart, Germany, 1977.